minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Config] Round 5.1: the published-side readers ask the bags, and a platform fact gets one address (#37086)

Browse Source
This commit is contained in:
Cheng Wan
2026-08-30 02:18:33 -07:00
committed by GitHub
parent a6e4021368
commit 7e751153eb
148 changed files with 1585 additions and 1114 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/unit/layers/quantization/test_compressed_tensors_wna16_moe_no_linear.py
+5 -12
View File
@@ -12,6 +12,7 @@ from sglang.srt.layers.quantization.compressed_tensors.schemes import (
CompressedTensorsWNA16MoE,
CompressedTensorsWNA16TritonMoE,
)
from sglang.srt.runtime_context import override_platform
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
@@ -90,9 +91,7 @@ class TestWNA16MoENoLinearGroup(CustomTestCase):
"get_moe_runner_backend",
return_value=MoeRunnerBackend.AUTO,
),
mock.patch.object(
compressed_tensors, "is_sm100_supported", return_value=True
),
override_platform(is_sm100=True),
):
scheme = quant_config.get_moe_scheme(
torch.nn.Module(), layer_name=EXPERTS_LAYER
@@ -120,9 +119,7 @@ class TestWNA16MoENoLinearGroup(CustomTestCase):
"get_moe_runner_backend",
return_value=MoeRunnerBackend.AUTO,
),
mock.patch.object(
compressed_tensors, "is_sm100_supported", return_value=True
),
override_platform(is_sm100=True),
):
scheme = quant_config.get_moe_scheme(
torch.nn.Module(), layer_name=EXPERTS_LAYER
@@ -157,9 +154,7 @@ class TestWNA16MoENoLinearGroup(CustomTestCase):
"get_moe_runner_backend",
return_value=MoeRunnerBackend.MARLIN,
),
mock.patch.object(
compressed_tensors, "is_sm100_supported", return_value=True
),
override_platform(is_sm100=True),
):
scheme = quant_config.get_moe_scheme(
torch.nn.Module(), layer_name=EXPERTS_LAYER
@@ -178,9 +173,7 @@ class TestWNA16MoENoLinearGroup(CustomTestCase):
"get_moe_runner_backend",
return_value=MoeRunnerBackend.AUTO,
),
mock.patch.object(
compressed_tensors, "is_sm100_supported", return_value=True
),
override_platform(is_sm100=True),
):
scheme = quant_config.get_moe_scheme(
torch.nn.Module(), layer_name=EXPERTS_LAYER
test/registered/unit/layers/quantization/test_mxfp4_sm120_cutlass.py
+5 -2
View File
@@ -11,6 +11,7 @@ from types import SimpleNamespace
import pytest
import torch
from sglang.srt.runtime_context import override_platform
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=120, stage="base-b", runner_config="1-gpu-small")
@@ -81,10 +82,12 @@ def test_cutlass_adapter_import_does_not_require_flashinfer(monkeypatch):
sys.modules[module_name] = cached_module
def test_dsv4_sm120_load_contract(monkeypatch):
def test_dsv4_sm120_load_contract(monkeypatch, request):
import sglang.srt.layers.quantization.mxfp4_flashinfer_cutlass_moe as adapter_module
monkeypatch.setattr(adapter_module, "is_sm120_supported", lambda: True)
platform = override_platform(is_sm120=True)
platform.install()
request.addfinalizer(platform.restore)
captured = {}